Skip to content

Latest commit

 

History

History
52 lines (40 loc) · 1.39 KB

README.md

File metadata and controls

52 lines (40 loc) · 1.39 KB
Raw

Polus

Polus: Sentiment Analysis for Greek Text based on "Leveraging aspect-based sentiment prediction with textual features and document metadata" paper for SETN2020. A Bi-LSTM neural network with 2 inputs (text and emojis map) for 3-class sentiment classification.

python version: Python 3.6.9

Clone Repo

git clone https://github.com/kkorovesis/polus.git

Install dependencies

pip install -r requirements.txt

Download embeddings

aws s3 cp s3://com.palo.ml-dev/embeddings/fasttext/cc.el.300.vec.gz embeddings/cc.el.300.vec.gz
gzip -d embeddings/cc.el.300.vec.gz
mv embeddings/cc.el.300.vec embeddings/fasttext.300d.txt

Train

Train one of the following models:

  • Polus TE model (train on text and emojis) -m polus_te
  • Polus TO model (train on text only) -m polus_to

config.ini: default params

python polus.py -d <train_data_filename> -m <polus_model> -c ./config.ini -f <model_filename> train --epochs <num_epochs> --batch <batch_size>

example:

python polus.py -d train_data.json -m polus_te -c ./config.ini -f te_model_ train --epochs <num_epochs> --batch <batch_size>

Test

python polus.py -d <train_data_filename> -m palo_te -c ./config.ini -f test

example:

python polus.py -d test_data.json -m palo_te -c ./config.ini -f test

Todos

[ ] Update dependencies

[ ] Include predict